Process for modelling a 3D scene

Abstract

A process for modeling a 3D scene is provided which comprises validating the model by determining a maximum permitted distortion ( 5 ) on a 2D synthesized image ( 4 ), generated by the approximation inherent in the model ( 2 ); calculating for a point I of a reference image ( 7 ) and on a set of synthesized images ( 4 ) representing the 3D point of the scene corresponding to this point I, of the minimum (z i −Δ zi 1 ) and maximum (z i +Δ zi 2 ) depth values of the depth z i of this point I corresponding to this maximum distortion, calculating a span around the depth z i of this point I, dependant on the minimum value of the error Δ zi 2 and on the minimum value of the error Δ zi 1 among the values calculated for the synthesized images of the set.

1. Process for modelling a 3D scene ( 1 ) defined on the basis of reference images ( 7 ) associated with viewpoints (I) and original depth maps, by calculation of approximate depth maps describing a model, comprising the following steps:
 predetermining of a maximum permitted distortion ( 5 ) on a 2D synthesized image ( 4 ), generated by the approximation inherent in the model ( 2 ), 
 calculating, for a point I of a reference image ( 7 ) and on the basis of a set of synthesized images ( 4 ) representing the 3D point of the scene corresponding to this point I, of the minimum (z i −Δ zi   1 ) and maximum (z i +Δ zi   2 ) depth values of a depth z i  of the point I corresponding to the maximum permitted distortion, 
 calculating a span ([z m (i), z M (i)]) around the depth z i  of this point I, dependant on the minimum value of the error Δ zi   2  and on the minimum value of the error Δ zi   1  among the values calculated for the synthesized images of the set, 
 and validating an approximate depth map calculated by iteration, as a function of the approximate depth z app  of 3D points of the corresponding model which relate to points I of the reference image, depending on whether it does or does not belong to the span ([z m (i), z M (i)]) calculated for these points I. 
 
   
   
     2. Process according to  claim 1 , wherein the calculation of the minimum and maximum values of the depth of a point I, for a synthesized image ( 4 ), is performed by utilizing an epipolar line corresponding to the point I in this synthesized image, to which are referred the distances corresponding to the maximum permitted distortion. 
   
   
     3. Process according to  claim 1 , wherein, the model is defined by a set of elementary regions, and the 3D points whose depth is utilized for the validating step are points of an elementary region of the model. 
   
   
     4. Process according to  claim 3 , wherein an elementary region is a facet and the calculation by iteration is a facetization of the depth maps, the iteration comprising creation of a new 3D facet vertex of the model or the deletion of a 3D facet vertex of the model depending on whether the process is of the “coarse to fine” or “fine to coarse” type. 
   
   
     5. Process according to  claim 4 , wherein if the process is of the “coarse to fine” type, the iteration comprises a calculation of the poorest approximate value from among the approximate values which relate to a facet and which are outside the spans, and the new vertex relates to a point corresponding to this value. 
   
   
     6. Process for compressing data defining a 3D scene, these data relating to reference images in respect of rendition and to associated depth maps, wherein the depth maps are modelled by facetization according to the process of  claim 4  and the data relating to the reference images and to the depth maps are replaced with data defining the facets. 
   
   
     7. Process for compressing data defining a 3D scene, these data relating to reference images in respect of rendition and to associated depth maps, comprising the modelling steps according to the modelling process of  claim 3 , the elementary regions of the model being defined by a partitioning of the depth maps into image blocks, the calculation by iterations on an approximate depth map consisting of a modification of a quantization step size for these blocks so as to give a quantized depth which is the approximate depth. 
   
   
     8. Process according to  claim 7 , wherein the modification of the quantization step size is an increase in the quantization step size to obtain the widest quantization step size or the same quantized depth values for a block. 
   
   
     9. Process according to  claim 1 , wherein the set of synthesized images ( 4 ) on which the calculation of the errors is performed corresponds to a sample representative of the set of viewpoints using the reference image ( 7 ). 
   
   
     10. Process according to  claim 9 , wherein the set of synthesized images on which the calculation of the errors is performed is composed of images having a number greater than a threshold of 3D points in common with the reference image. 
   
   
     11. Process according to  claim 1 , wherein the maximum permitted distortion is predetermined as a function of a characteristic of an image. 
   
   
     12. Process according to  claim 11 , wherein a characteristic of an image is the luminance gradient. 
   
   
     13. Process for generating images, wherein the images are generated on the basis of a modelling of the scene according to the process of  claim 1 .